Abstract
<title>Abstract</title> <p>Causal reasoning, inferring what leads to what, is the foundation upon which animals and humans build their understanding of the world. Yet causal relations are rarely fully revealed; they must be inferred from incomplete and ambiguous observations. We propose that such causal inference operates through the integration of two complementary models of the environment: a relational structure, encoding cause-effect relationships, and an abstract world model, encoding the generative principles that govern how causal relations are formed and interpreted. To dissociate these two processes, we developed a novel one-shot causal inference paradigm. Combining both neural recording and intervention methods, we reveal that the two proposed models are neurally dissociable yet functionally coordinated: the hippocampus instantiates the relational structure by building a map of causal relations, a process controlled by the inferior frontal sulcus (IFS), which is critically involved in inferring unobserved causal relations through its encoding of the abstract world model.</p>